This whitepaper details a production‑grade, multi‑layer anti‑fraud system that starts with low‑cost traffic filtering, builds rich risk signals, orchestrates rules, models and graph analysis for real‑time decisions, and embeds governance, scalability and high‑concurrency practices to protect online services.
The Feature Center in iQIYI’s Opal platform centralizes feature creation, storage, and real‑time access through a drag‑and‑drop DAG workflow and DSL‑driven transformations, handling massive QPS and low‑latency demands while enabling fast business iteration, cross‑team reuse, and monitoring for advertising, recommendation, and risk‑control applications.
This article details Alipay's advertising algorithm upgrades—including sample‑enhanced cold‑start mitigation, cross‑scene and user‑segmented knowledge transfer, and real‑time feature and online‑learning optimizations—that collectively lifted CTR, CPM, and overall business revenue.
This article presents OPPO’s next‑generation big‑data and AI integrated architecture on functional cloud, detailing a cloud‑native elastic compute framework, a unified data‑lake solution, real‑time feature platforms, machine‑learning data acceleration, and hybrid‑cloud deployments, highlighting performance gains and cost reductions.
The article describes how a Flink‑based streaming prediction platform was built to flatten peak request loads, reduce latency and memory use, and improve stability by deduplicating SDK calls, incrementally loading Hive features, partitioned caching, and comprehensive monitoring, while a templating system automates feature definition, SQL generation and stress testing, enabling real‑time supply‑demand forecasting that outperforms offline methods.
This article presents ZhongAn Financial’s end‑to‑end MLOps workflow and real‑time feature platform architecture, detailing team roles, data pipelines, Flink‑based processing, TableStore storage, anti‑fraud feature design, and answers to common implementation questions, offering a comprehensive guide for building scalable, low‑latency ML services in finance.
The article reviews the development of NetEase Leihuo's game recommendation system, covering the shift from offline batch recommendation to real‑time feature engineering and online inference, detailing architecture design, practical experiences, performance optimizations, and future directions such as real‑time training.
The team built a lightweight AI‑FAAS platform that migrates C++ ad‑algorithm services to a Java‑centric, cloud‑native environment, encapsulates core operators as DolphinSQL plugins, and defines real‑time feature pipelines via standardized SQL, cutting service creation from weeks to days, enabling rapid iteration, high operator reuse, and near‑instant feature delivery.
The article describes a unified global feature pool architecture that standardizes offline and real‑time feature production, management, and service layers using Hive, Spark, Flink, Kafka, MySQL, and Hologres to break data silos, improve algorithm development efficiency, and boost growth business performance.
Peng Mo’s talk details Tencent Kankan’s billion‑user feed architecture—layered data, recall, ranking, and exposure control—while addressing real‑time feature generation, massive concurrency, memory‑intensive caching, and fast indexing, and explains solutions such as multi‑level caches, online minute‑level model updates, Redis bloom‑filter exposure filtering, a lock‑free hash‑plus‑linked‑list index, and distributed optimizations that halve latency to under 500 ms and support auto‑scaling and cold‑start handling.
Meituan’s Turing platform consolidates the entire delivery‑order workflow—from massive data ingestion and feature generation to model training, evaluation, deployment, real‑time prediction, and AB testing—into a single, end‑to‑end system that evolved from a minimal MVP into a fully platformized solution, addressing speed, accuracy, and engineering‑algorithm decoupling while planning deeper deep‑learning integration.
